Don't like the DSP function. Wait, what?

To my very great surprise, I don’t much like the DSP function.
I really expected to use it and love it. Have been waiting for it. I have wanted room correction for some time but never got around to it. Now I could do it the server, for all my rooms and endpoints. Hooray!

But I don’t like it. Maybe when I get better at it. I’ll keep trying.

My preference is for clarity and precision, over most other attributes. (Not all, more on this later.) My gear is chosen accordingly.

So I experimented with room correction, Parametric and Procedural and Convolution, with different targets, house curves, etc. Used REW, and trial-and-error. The results measured much flatter, although not perfect. But it sounded variously thin, or muddy, or harsh, depending on how I defined the filters. Every time it was a relief to turn them off.

Was I just used to the old sound? Possible, but I gave it an honest try.

Is it just my incompetence? Quite possible. I have no experience with this. I well remember Malcolm Gladwell’s “10,000 hours to mastery”. And I didn’t have the best software tools (although REW is well regarded, Acourate may be better but costs nearly as much as Roon). But is the average user going to be more skilled than I am, or have better tools? I am going to get Acourate and try it, just because I’m a geek and I love a challenge. But so far…

For my headphones, I tried crossfeed. Muddy. Lost the all-important clarity. A relief to restore the original sound. Maybe if I had worse recordings…

Upsampling: DSD is not much value for the Meridian gear which runs on PCM internally, for crossovers and DSP. Upsampling the PCM didn’t do anything, Meridian is already good at it. Plus it interferes with MQA (so far). The Geek Pulse headphone amp accepts DSD128, but upsampling in Roon wasn’t better than letting the Geek do it, they are good at it; initially tempting, but eventually unsatisfactory.

Are all these special circumstances with my gear? Or incompetence? Or comfort with the old sound? Or my stubborn preference for clarity?

I don’t know. Will continue experimenting.

It would be interesting to hear others’ experiences.


On clarity, see my note in the Music section.

Roon’s DSP fixed a problem for me…

In the den (TV Room) I share with girlfriend/fine companion, there’s a 60" TV with cable box and Apple TV for streaming. For audio we use a soundbar with a wireless subwoofer. The sub sits in a corner due to GAF. The low frequency boominess of this setup doesn’t bother me when watching TV but it really prevented me from enjoying music streaming via Roon in that room to the Apple TV.

I used Roon’s DSP feature and enabled the Parametric EQ with a high shelf filter for that zone. (Apple TV-Den) I futzed with the settings a little bit to get a flat response and I must say I am very pleased with the SQ in that room now. Given that it’s a soundbar via Airplay my expectations aren’t exactly high but now I can sit and listen to music with GF. Simple fix.

I barely glanced at the Roon knowledge-base info on using DSP beforehand and I think the whole process took me 15 minutes tops. The interface is simple but the results, in my opinion, are quite sophisticated.

In my main listening room, I have Roon’s DSP engine set up for various upsampled rates as I browse the values to see if I can hear a difference. Very nice but no conclusions yet. (Maybe we need a DSP setting that we can set for each album?)

Haven’t tried anything more than that yet but so far I am greatly enjoying this new feature.

[quote=“AndersVinberg, post:1, topic:21116”]
So I experimented with room correction, Parametric and Procedural and Convolution, with different targets, house curves, etc. Used REW, and trial-and-error. The results measured much flatter, although not perfect. But it sounded variously thin, or muddy, or harsh, depending on how I defined the filters. Every time it was a relief to turn them off.
[/quote]have a read of I’ve heard great results using DRC, but principally to deal with LF issues rather than trying to address the whole audible spectrum.

I just did room correction in a small library, with small but good speakers, symmetrical. And it worked well, took care of some room modes. +/- 3 dB 20-20,000, with some dips and one spike. And it sounds good too.

Limited ambitions first, it seems.

For me, DSP has been a positive development.
I have several rather modest systems (5 zones) throughout the house, all served by my Roon Core.

Upsampling has smoothed out SQ in every system. I expected this, it’s the same result I got when I played with my trial of HQPlayer last fall. I’m very glad it’s now available from within Roon, and have enjoyed the results.

Working with the Speaker Setup options has brought good results too. Unfortunately, my listening situations are less than perfect due to room shape, required furniture placement, materials, etc. I don’t have a perfectly tailored, dedicated listening room (wish I did!) so I’ve worked with the controls in this section a lot in each room. I’ve had much better results at improving soundstage, depth, etc. with Roon’s DSP than with other options I’ve tried. Each space now feels much more natural and balanced than before, and the sound is actually much the same from space to space.

Parametric EQ is more problematic, due I believe, to the learning curve. I took big swings in each space at first, and was underwhelmed with the results. The sound was uneven and muddy. I started over, working in smaller increments, making subtler changes, and finally found curves that have definitely improved SQ relative to leaving the feature off. So far, so good - it’s still on in each zone. I love having independently saved settings for each zone.

My results may be skewed by the mid-fi nature of my various systems. I love vintage stereo, and have rebuilt systems from various eras. Tube amps from the 50’s and 60’s, silver-faced receivers from the seventies - all refurbished from the ground up. Speakers by Klipsch, DCM, and B&W. None of my equipment (except the digital bits) would be considered modern or high-end. As would be expected, there is a lot of variation to dial in. The new tools in Roon have been fun to work with, and have shown good results. I’m pleased thus far - thanks to the team at Roon!

My experience with the new DSP features have been similar to Bill’s:

  1. Speaker correction–I think it really did make an audible improvement as my listening “sweet spot” is back further than recommended due to furniture/room issues. The millisecond changes combined w/my SPL meter measurements to boost one speaker by 2-3 dB was a noticeable improvement.
  2. Upsampling–I think I can tell an improvement w/full PCM upsampling to 384 for PCM; leaving my native DSD files alone. Unfortunately, my MacMini is a traditional harddrive not SSD so I was getting 20-30 x w/PCM upsampling but only 1.2-1.5 x w/DSD upsampling along w/stutters occ at DSD128 and unlistenable stutters or no sound at all with DSD256 (my iFiMicro DSD will accept up to DSD512). Plus even when I could eek out a DSD upsample rate of 3-5 x, I still just didn’t like the sound. I love DSD and have literally more than a thousand SACDs, but prob need more computer horsepower to do the Roon DSD upsampling justice. Not willing to buy a new computer at the moment…Apple needs to come out w/an updated Mac Mini, then I will bite.
  3. Convolution–I haven’t tried REW or Accourate, but I did download the free trial of Dirac and made several sweeps w/“optimization”. Since I can’t upload Dirac files into Roon as zip files and didn’t want to use it as a play through zone, I attempted to recreate the curves from Dirac in the parametric EQ in Roon by just writing down the freq/gains from the optimized Dirac sweeps. Still didn’t really love it and some recreated curves had some pretty good clipping tamed w/the Headroom Function. Still wanna try REW since I have the miniDSP mic.
  4. Parametric EQ–f/u to #3, I then read more online about Parametric EQ’s and simply fiddled w/the settings. I still used the Dirac recreated curves but significantly decreased the gain of each and played around w/slope a bit. A few freq frankly sounded better going in the opposite direction on the gain from Dirac, others were spot on.

What do you all think of that as a poor man’s convolution? What else will REW do besides what I have done so far? I do think the tweaked Parametric EQ sounds significantly better “on” vs “off” and better than the preset “flat” (which honestly sounded better “on” than “off” in my system though still not quite as good as the tweaked version with a slight inc in base freq and slow roll off in high freq).

There’s definitely a “less is more” aspect to DSP. It’s very easy to do too much and make things worse.

First and foremost, I use DSP for experience benefits. I use volume leveling in nearly every zone. In a couple of zones, I also normalize everything to one sample rate to limit the amount of switching going on. An incredible number of DACs out there are not perfectly clean during transitions, and it’s distracting.

EQ’ing a room is tough, but EQ’ing headphones is easier–because someone out there with better measuring equipment has already done the measurements and worked out some parameters. PEQ curves for every popular headphone model are easily googleable and there are usually plenty of options to play with.

One DAC that I’ve spent a lot of time with has a separate hardware path for DSD–so DSD upsampling is a mechanical requirement for using that half of the hardware.

Upsampling is a little bit trickier. As you found with the Meridian stuff, the differences between upsampling filters that are pointed at the same set of goals (minimum phase apodising filters, for example) are subtle.

That said, if you believe that minimum phase apodising filters provide a benefit–then there is some potential to improve onthe LH Labs DAC. As far as I can remember, that one is a pair of ES9018K2M/AQ2M chipdacs in a box with no discrete upsampling stage. Those chips have programmable FIR filters, but the coefficient budget is too constrained to do what Meridian’s or Roon’s converters are doing.

On the other hand, I spent a few minutes adjusting the positioning of the speakers around my desk the other day and did more good than 4 hours of fiddling around in REW with the speakers in the old position.


Same here. I absolutely love Roon DSP’s usability and flexibility, but honestly, I don’t particularly like how it sounds. For some reason it seems to sound a bit veiled, not matter what settings I have there. I get better sound quality either by disabling the Roon’s DSP or by using the HQPlayer. Fortunately I can use both software together, so I’m happy :slight_smile:

1 Like


I’m playing for a couple hours with the DSP an each time I prefer without it.
I was mainly interested about Tidal upsampling. (Not MQA)
I cannot find the words to explain the rationale. I’m more captured by the music without oversampling.
My current system is a (quite old 4Gb) Mac Mini connected to a Devialet 200 via USB.
Should I prefer PCM over DSD because of the Devialet? I tried all combination and I always go back to disabling upsampling.

I’m about to buy a Vortexbox and a Microrendu to replace the Mini. I don’t know if it will make any difference.



After more experience, I’m now more fond of upsampling for the LHLabs Pulse.

And I am experimenting with PEQ for the LCD-3. As you say, Audeze provides a measurement curve, they are very flat except for a severe dropout at 2,000 - 10,000, I corrected that and I’m evaluating it now.

For room EQ, I wasn’t pleased with what I got with REW, have just bought Acourate but not had time to work with it. But I think there is much to be done, the REW and XTZ measurements show a lot of room effects.

We keep learning.

The crossfeed is muddy, but it makes many poorly mixed songs listenable with headphones. My favorite example is Maggot Brain by Funkadelic. It’s one of my favorite songs, but the left right panning tricks make it a circus in my head. Crossfeed reduces the clarity but makes it listenable. I don’t use it often, but it’s another item in the toolbox to make the music more enjoyable.

We are working on better crossfeed solutions.


See what you mean, but a work called “Maggot Brain” by Funkadelic – you’ve actually got to be kidding, right?

Listen to it… You won’t be disappointed.

According to legend, George Clinton, under the influence of LSD, told Eddie Hazel during the recording session to imagine he had been told his mother was dead, but then learned that it was not true. The result was the 10-minute guitar solo for which Hazel is most fondly remembered by many music critics and fans. Though several other musicians began the track playing, Clinton soon realized how powerful Hazel’s solo was and faded them out so that the focus would be on Hazel’s guitar. Critics have described the solo as “lengthy, mind-melting” and “an emotional apocalypse of sound.”

Another really emotional guitar piece is Soothsayer by Buckethead.


It’s totally unlike the rest of George Clinton’s work, very unique, and very special.

1 Like

You are right. Good to be yanked out of one’s preconceptions. I did enjoy this stuff a long time ago, but forgot. Thanks.

Btw, I’m on vacation, sitting on the lanai in a Hawaii resort, and (as we have discussed many times) I don’t have Roon with me. But I do have Tidal, and the Audeze Lightning-connected DAC does a good job of feeding Maggot Brain into the EL-8s. Can’t wait to feed it into my main rig.

@danny @brian In fact, I think I just made the overwhelming case why mobile is a more important next feature than all the others.


Big fan of “Maggot Brain” here as well - but writing to ask some more information on DSP volume control: how does it work “technically” ? Does it degrade the sound in any way ? What are others’ experience using it ?


I also use a Devialet - 250 Pro - and I find the upsampling in the Devialet to be extremely good. Most of the time I prefer to input native sample rate to the Dev and let it do the job. I do not have any DSD files so I can not comment on that. I am very happy with the convolution engine in Roon. I’ve used RC for many years and it has never been easier to use than this.

What do you use to measure and generate the convolution?